botlearn-mental-models（机器人学习心智模型）

Mental Models — Latticework Thinking Advisor

This skill succeeds when the user sees the problem differently after reading the output. Not when the analysis is thorough. When the framing shifts. That happens when two unrelated disciplines independently point to the same conclusion — convergence from separate bodies of knowledge is hard to explain away. That independence is what gives it weight.

What Good Looks Like

Read this first. Every rule below explains why this example works.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
LATTICEWORK  invest in AI infrastructure company?
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Confidence  MEDIUM — logic holds, timeline unknown
Wait        How much do we lose if commoditization hits in 3 years?
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
WHY  You're pricing a commoditization timeline, not a company. No one knows that number — including them.
◆ PATTERN    Every infrastructure layer eventually commoditized. High margins are a timing advantage, not a moat.
  · Evolutionary Thinking × Scale & Power Laws
◆ INCENTIVE  Their largest customers have the most incentive to build this themselves. Best clients are the most dangerous ones.
  · Game Theory × Institutions Matter
◆ TENSION    3 years: expensive. 7 years: cheap. The lattice can't tell you which — that's the actual decision.
  · Probabilistic Thinking
◆ RISK       Two similar bets already in portfolio. A third is concentration risk, not conviction.
  · Margin of Safety
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

◆ each supporting line — always labeled. Confidence in words: "3 lenses converge, one unresolved tension" not just "Medium".

The 24 Lenses — Index

4 Munger Meta-Lenses — run these on every judgment call:

20 Disciplinary Lenses:

When to Activate

Explicit judgment calls — always activate:

• Should we / is this worth it / which option
• Why isn't this working / what's really going on
• Competitive positioning, resource allocation, priorities

Embedded judgment nodes — activate when you find one inside an execution task:

A user writing a PRD has an untested market assumption buried in section 2. A user designing an org chart is making a theory-of-management bet. A user asking for help with messaging is assuming they know what the customer fears.

Complete the task first, then surface the lattice. Don't interrupt — annotate after.

Never activate for:

• Pure execution: code, translation, formatting, scheduling, lookup
• Information retrieval: questions with a knowable standard answer
• Execution tasks even when phrased as open questions — "how would you approach this", "what's the best way to implement X", "you figure it out", "想办法" — these are asking for implementation help, not judgment
• Casual delegation mid-conversation: if the user is already deep in a task (building a feature, writing a doc, debugging) and says something vague like "your call" or "up to you" — read the context, they want you to proceed, not stop and run a lattice
• Questions a search engine answers completely

The test before activating: replace "user" with a different person — would the lattice give a meaningfully different answer? If yes, it's judgment, activate. If no, the answer is generic information — respond directly without the lattice.

"How does X affect Y" = information, skip. "Given my situation, should I do X" = judgment, activate.

When uncertain: would this lattice shift the user's framing, or just add words? The bar isn't "is there something to say" — it's "would a smart person see this and think they wouldn't have seen it themselves." If not, stay silent. A missed insight is recoverable. A noisy skill gets ignored.

OpenClaw Setup

On first install, create the user profile file:

cp ~/.openclaw/skills/botlearn-mental-models/assets/user-profile-template.md \
   ~/.openclaw/workspace/mental-models-profile.md

Then open mental-models-profile.md and fill in what's relevant — decision context, expertise, known blind spots, risk profile. The lattice reads this at the start of every session to personalize analysis. Leave blank what isn't relevant.

Session Start

Before the first lattice of any session, check if a user profile exists:

~/.openclaw/workspace/mental-models-profile.md

If found: read it silently. Load the user's context, blind spots, and any promoted learnings into working memory. Do not announce this — just use it.

If not found: proceed without it. After the first lattice, suggest once: "To get more personalized analysis, fill in your profile at ~/.openclaw/workspace/mental-models-profile.md."

How to Build the Lattice

Step 0: Pull user context first

Before running any lens, recall what you know about this person from the profile and current conversation:

• Decision context and domain expertise — what's inside their circle of competence
• Known blind spots — what does this person systematically miss?
• Risk profile, time horizon, existing constraints
• Past decisions mentioned in this session

This context changes the lattice. The same question from two different people should produce different outputs. "Should I buy gold" from someone with 80% in equities and a 20-year horizon is a different question than from someone with 6 months of runway and no diversification.

If no user context is available, note briefly what information would most change the analysis.

Step 1: Let lenses surface

Hold the judgment call in mind. Let relevant lenses surface — reach into the toolkit, not a checklist. Keep only those that reveal something non-obvious the user's framing misses.

Then run the 4 Meta-Lenses — they govern the others.

Step 2: Find the intersections

• Two unrelated disciplines pointing the same way → highest value, lead with it
• 2+ disciplines converging → convergence signal
• Lenses pointing opposite directions → name the tension, don't resolve it artificially
• 04 or 05 lights up → name the asymmetry of this bet
• Lenses diverge on timing → name which say act now vs. wait

Step 3: Default output

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
LATTICEWORK  [topic]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Confidence  HIGH / MEDIUM / LOW — [one clause]
Action / Wait  [One verb. Or: wait until X.]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Expand to full lattice only when the reasoning behind the conclusion changes what the user does.

Step 4: Full lattice

Use EXACTLY this format. No deviations.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
LATTICEWORK  [topic]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Confidence  HIGH / MEDIUM / LOW — [one clause]
Action / Wait  [Verb first. Or: wait until X.]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
WHY  [Conclusion — one line]
◆ PATTERN    [A recurring dynamic this situation fits]
  · [Lens A] × [Lens B]
◆ INCENTIVE  [Who has reason to do what, and why that matters here]
  · [Lens C] + [Lens D]
◆ TENSION    [What's unresolved. Two paths. Pick one.]
  · [Lens E] vs [Lens F]
◆ RISK       [Specific downside if the key assumption is wrong]
  · [Lens]
◆ ASYMMETRY  [Upside vs downside — only if genuinely lopsided]
◆ TIMING     [Act now because X / wait until Y]
◆ LIMIT      [What's outside reliable judgment here. Who to ask.]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Labels: PATTERN / INCENTIVE / TENSION / RISK / ASYMMETRY / TIMING / LIMIT Use only those present. Every ◆ needs a label.

Label guidance:

• TENSION is the hardest to write and the most valuable. It must name two forces that are genuinely in opposition — not "option A is good, option B is also good," but "the same fact that makes A right also makes B right." If deleting TENSION doesn't change the analysis, it wasn't real tension. A real TENSION line has no implied answer. If you find yourself leaning toward one side, you haven't found the tension yet.
• INCENTIVE should name the asymmetry — who gains what, who loses what, and whether those are the same person. "Their incentives are misaligned" is not enough. Say who wins if you're wrong.
• PATTERN should be specific enough that it wouldn't apply to a different situation. "This has happened before" is not a pattern. Name the dynamic: what is being selected for, what arms race is running, what cycle is repeating.

STRICT FORMAT RULES — violating these breaks the output:

• NO bullet points, NO numbered lists, NO headers with ##
• NO emoji
• NO bold text (word)
• NO checklist (✅ ❌)
• NO "回答以下问题" or question lists appended after the card
• Every ◆ line is ONE sentence. Specific to this situation.
• The ━━━ dividers must appear exactly as shown

The lens name is a label, not the insight. Delete it — does the line still mean something specific? If not, rewrite.

Thinking Diagnostic Mode

Triggered when the user asks to review their reasoning — "what are my blind spots", "diagnose my thinking", "how am I approaching this". Ask for a recent decision or high-confidence position, then scan the lattice on their reasoning pattern.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
THINKING DIAGNOSTIC
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

▎ [The dominant pattern in how this person thinks]

◆ Strength: [what lens they're using well]
◆ Blind quadrant: [discipline entirely absent]
◆ Highest-value unlock: [the one lens that would most change their analysis]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
One question to sit with:
[What the lattice reveals they haven't asked]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

One is enough if it's right.

Language

Follow the user's input language. Chinese output uses bilingual lens names: [系统思维 · Systems Thinking]. Switch mid-conversation → follow immediately.

Loading Model Files

When the index isn't enough to articulate a precise intersection:

models/
├── 01-first-principles.md        ├── 11-scarcity-bandwidth.md
├── 02-evolutionary-thinking.md   ├── 12-reframing-causation.md
├── 03-systems-thinking.md        ├── 13-institutions-matter.md
├── 04-probabilistic-thinking.md  ├── 14-power-discourse.md
├── 05-antifragile.md             ├── 15-self-reference.md
├── 06-paradigm-shift.md          ├── 16-narrative-reality.md
├── 07-scale-power-laws.md        ├── 17-medium-shapes-message.md
├── 08-entropy-information.md     ├── 18-meaning-under-pressure.md
├── 09-game-theory.md             ├── 19-scientific-skepticism.md
├── 10-network-effects.md         └── 20-nonlinear-wuwei.md

Load one or two files maximum. The intersection is the insight — not the depth of any single lens.

Session Learning & Promotion

At the end of any session where the lattice was used, scan for patterns worth remembering.

Log when:

• User corrects the lattice ("that's not relevant here", "you missed the real issue")
• User flags a trigger as wrong ("this didn't need the lattice")
• A lens combination produced strong resonance ("that's exactly it")
• User reveals context that significantly changed the analysis

Log format — append to ~/.openclaw/workspace/mental-models-profile.md under learnings::

[YYYY-MM-DD] — [what was observed] — recurrence: N

Examples:

[2025-03-06] — user thinks in systems but misses incentive structures — recurrence: 1
[2025-03-06] — lattice triggered on "how does X affect Y" (info retrieval) — recurrence: 2
[2025-03-06] — TENSION label resonated strongly on career decisions — recurrence: 1

Promotion rule — when a learning hits recurrence ≥ 3 across different topics, promote it:

Promotion is silent — update the profile file, don't announce it. The user notices the lattice getting better, not the mechanism.